xref: /freebsd/sys/kern/kern_exit.c (revision e2eeea75eb8b6dd50c1298067a0655880d186734)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/capsicum.h>
48 #include <sys/eventhandler.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.h>
51 #include <sys/malloc.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/procdesc.h>
56 #include <sys/jail.h>
57 #include <sys/tty.h>
58 #include <sys/wait.h>
59 #include <sys/vmmeter.h>
60 #include <sys/vnode.h>
61 #include <sys/racct.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sbuf.h>
64 #include <sys/signalvar.h>
65 #include <sys/sched.h>
66 #include <sys/sx.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/syslog.h>
69 #include <sys/ptrace.h>
70 #include <sys/acct.h>		/* for acct_process() function prototype */
71 #include <sys/filedesc.h>
72 #include <sys/sdt.h>
73 #include <sys/shm.h>
74 #include <sys/sem.h>
75 #include <sys/umtx.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 
80 #include <security/audit/audit.h>
81 #include <security/mac/mac_framework.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/uma.h>
90 
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t	dtrace_fasttrap_exit;
94 #endif
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, , , exit, "int");
98 
99 struct proc *
100 proc_realparent(struct proc *child)
101 {
102 	struct proc *p, *parent;
103 
104 	sx_assert(&proctree_lock, SX_LOCKED);
105 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
106 		return (child->p_pptr->p_pid == child->p_oppid ?
107 		    child->p_pptr : child->p_reaper);
108 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
109 		/* Cannot use LIST_PREV(), since the list head is not known. */
110 		p = __containerof(p->p_orphan.le_prev, struct proc,
111 		    p_orphan.le_next);
112 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
113 		    ("missing P_ORPHAN %p", p));
114 	}
115 	parent = __containerof(p->p_orphan.le_prev, struct proc,
116 	    p_orphans.lh_first);
117 	return (parent);
118 }
119 
120 void
121 reaper_abandon_children(struct proc *p, bool exiting)
122 {
123 	struct proc *p1, *p2, *ptmp;
124 
125 	sx_assert(&proctree_lock, SX_LOCKED);
126 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
127 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
128 		return;
129 	p1 = p->p_reaper;
130 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
131 		LIST_REMOVE(p2, p_reapsibling);
132 		p2->p_reaper = p1;
133 		p2->p_reapsubtree = p->p_reapsubtree;
134 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
135 		if (exiting && p2->p_pptr == p) {
136 			PROC_LOCK(p2);
137 			proc_reparent(p2, p1, true);
138 			PROC_UNLOCK(p2);
139 		}
140 	}
141 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
142 	p->p_treeflag &= ~P_TREE_REAPER;
143 }
144 
145 static void
146 reaper_clear(struct proc *p)
147 {
148 	struct proc *p1;
149 	bool clear;
150 
151 	sx_assert(&proctree_lock, SX_LOCKED);
152 	LIST_REMOVE(p, p_reapsibling);
153 	if (p->p_reapsubtree == 1)
154 		return;
155 	clear = true;
156 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
157 		if (p1->p_reapsubtree == p->p_reapsubtree) {
158 			clear = false;
159 			break;
160 		}
161 	}
162 	if (clear)
163 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
164 }
165 
166 void
167 proc_clear_orphan(struct proc *p)
168 {
169 	struct proc *p1;
170 
171 	sx_assert(&proctree_lock, SA_XLOCKED);
172 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
173 		return;
174 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
175 		p1 = LIST_NEXT(p, p_orphan);
176 		if (p1 != NULL)
177 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
178 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
179 	}
180 	LIST_REMOVE(p, p_orphan);
181 	p->p_treeflag &= ~P_TREE_ORPHANED;
182 }
183 
184 /*
185  * exit -- death of process.
186  */
187 void
188 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
189 {
190 
191 	exit1(td, uap->rval, 0);
192 	/* NOTREACHED */
193 }
194 
195 /*
196  * Exit: deallocate address space and other resources, change proc state to
197  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
198  * and rusage for wait().  Check for child processes and orphan them.
199  */
200 void
201 exit1(struct thread *td, int rval, int signo)
202 {
203 	struct proc *p, *nq, *q, *t;
204 	struct thread *tdt;
205 	ksiginfo_t *ksi, *ksi1;
206 	int signal_parent;
207 
208 	mtx_assert(&Giant, MA_NOTOWNED);
209 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
210 
211 	p = td->td_proc;
212 	/*
213 	 * XXX in case we're rebooting we just let init die in order to
214 	 * work around an unsolved stack overflow seen very late during
215 	 * shutdown on sparc64 when the gmirror worker process exists.
216 	 * XXX what to do now that sparc64 is gone... remove if?
217 	 */
218 	if (p == initproc && rebooting == 0) {
219 		printf("init died (signal %d, exit %d)\n", signo, rval);
220 		panic("Going nowhere without my init!");
221 	}
222 
223 	/*
224 	 * Deref SU mp, since the thread does not return to userspace.
225 	 */
226 	td_softdep_cleanup(td);
227 
228 	/*
229 	 * MUST abort all other threads before proceeding past here.
230 	 */
231 	PROC_LOCK(p);
232 	/*
233 	 * First check if some other thread or external request got
234 	 * here before us.  If so, act appropriately: exit or suspend.
235 	 * We must ensure that stop requests are handled before we set
236 	 * P_WEXIT.
237 	 */
238 	thread_suspend_check(0);
239 	while (p->p_flag & P_HADTHREADS) {
240 		/*
241 		 * Kill off the other threads. This requires
242 		 * some co-operation from other parts of the kernel
243 		 * so it may not be instantaneous.  With this state set
244 		 * any thread entering the kernel from userspace will
245 		 * thread_exit() in trap().  Any thread attempting to
246 		 * sleep will return immediately with EINTR or EWOULDBLOCK
247 		 * which will hopefully force them to back out to userland
248 		 * freeing resources as they go.  Any thread attempting
249 		 * to return to userland will thread_exit() from userret().
250 		 * thread_exit() will unsuspend us when the last of the
251 		 * other threads exits.
252 		 * If there is already a thread singler after resumption,
253 		 * calling thread_single will fail; in that case, we just
254 		 * re-check all suspension request, the thread should
255 		 * either be suspended there or exit.
256 		 */
257 		if (!thread_single(p, SINGLE_EXIT))
258 			/*
259 			 * All other activity in this process is now
260 			 * stopped.  Threading support has been turned
261 			 * off.
262 			 */
263 			break;
264 		/*
265 		 * Recheck for new stop or suspend requests which
266 		 * might appear while process lock was dropped in
267 		 * thread_single().
268 		 */
269 		thread_suspend_check(0);
270 	}
271 	KASSERT(p->p_numthreads == 1,
272 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
273 	racct_sub(p, RACCT_NTHR, 1);
274 
275 	/* Let event handler change exit status */
276 	p->p_xexit = rval;
277 	p->p_xsig = signo;
278 
279 	/*
280 	 * Ignore any pending request to stop due to a stop signal.
281 	 * Once P_WEXIT is set, future requests will be ignored as
282 	 * well.
283 	 */
284 	p->p_flag &= ~P_STOPPED_SIG;
285 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
286 
287 	/* Note that we are exiting. */
288 	p->p_flag |= P_WEXIT;
289 
290 	/*
291 	 * Wait for any processes that have a hold on our vmspace to
292 	 * release their reference.
293 	 */
294 	while (p->p_lock > 0)
295 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
296 
297 	PROC_UNLOCK(p);
298 	/* Drain the limit callout while we don't have the proc locked */
299 	callout_drain(&p->p_limco);
300 
301 #ifdef AUDIT
302 	/*
303 	 * The Sun BSM exit token contains two components: an exit status as
304 	 * passed to exit(), and a return value to indicate what sort of exit
305 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
306 	 * what the return value is.
307 	 */
308 	AUDIT_ARG_EXIT(rval, 0);
309 	AUDIT_SYSCALL_EXIT(0, td);
310 #endif
311 
312 	/* Are we a task leader with peers? */
313 	if (p->p_peers != NULL && p == p->p_leader) {
314 		mtx_lock(&ppeers_lock);
315 		q = p->p_peers;
316 		while (q != NULL) {
317 			PROC_LOCK(q);
318 			kern_psignal(q, SIGKILL);
319 			PROC_UNLOCK(q);
320 			q = q->p_peers;
321 		}
322 		while (p->p_peers != NULL)
323 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
324 		mtx_unlock(&ppeers_lock);
325 	}
326 
327 	/*
328 	 * Check if any loadable modules need anything done at process exit.
329 	 * E.g. SYSV IPC stuff.
330 	 * Event handler could change exit status.
331 	 * XXX what if one of these generates an error?
332 	 */
333 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
334 
335 	/*
336 	 * If parent is waiting for us to exit or exec,
337 	 * P_PPWAIT is set; we will wakeup the parent below.
338 	 */
339 	PROC_LOCK(p);
340 	stopprofclock(p);
341 	p->p_ptevents = 0;
342 
343 	/*
344 	 * Stop the real interval timer.  If the handler is currently
345 	 * executing, prevent it from rearming itself and let it finish.
346 	 */
347 	if (timevalisset(&p->p_realtimer.it_value) &&
348 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
349 		timevalclear(&p->p_realtimer.it_interval);
350 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
351 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
352 		    ("realtime timer is still armed"));
353 	}
354 
355 	PROC_UNLOCK(p);
356 
357 	umtx_thread_exit(td);
358 	seltdfini(td);
359 
360 	/*
361 	 * Reset any sigio structures pointing to us as a result of
362 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
363 	 */
364 	funsetownlst(&p->p_sigiolst);
365 
366 	/*
367 	 * Close open files and release open-file table.
368 	 * This may block!
369 	 */
370 	pdescfree(td);
371 	fdescfree(td);
372 
373 	/*
374 	 * If this thread tickled GEOM, we need to wait for the giggling to
375 	 * stop before we return to userland
376 	 */
377 	if (td->td_pflags & TDP_GEOM)
378 		g_waitidle();
379 
380 	/*
381 	 * Remove ourself from our leader's peer list and wake our leader.
382 	 */
383 	if (p->p_leader->p_peers != NULL) {
384 		mtx_lock(&ppeers_lock);
385 		if (p->p_leader->p_peers != NULL) {
386 			q = p->p_leader;
387 			while (q->p_peers != p)
388 				q = q->p_peers;
389 			q->p_peers = p->p_peers;
390 			wakeup(p->p_leader);
391 		}
392 		mtx_unlock(&ppeers_lock);
393 	}
394 
395 	vmspace_exit(td);
396 	(void)acct_process(td);
397 
398 #ifdef KTRACE
399 	ktrprocexit(td);
400 #endif
401 	/*
402 	 * Release reference to text vnode
403 	 */
404 	if (p->p_textvp != NULL) {
405 		vrele(p->p_textvp);
406 		p->p_textvp = NULL;
407 	}
408 
409 	/*
410 	 * Release our limits structure.
411 	 */
412 	lim_free(p->p_limit);
413 	p->p_limit = NULL;
414 
415 	tidhash_remove(td);
416 
417 	/*
418 	 * Call machine-dependent code to release any
419 	 * machine-dependent resources other than the address space.
420 	 * The address space is released by "vmspace_exitfree(p)" in
421 	 * vm_waitproc().
422 	 */
423 	cpu_exit(td);
424 
425 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
426 
427 	/*
428 	 * Remove from allproc. It still sits in the hash.
429 	 */
430 	sx_xlock(&allproc_lock);
431 	LIST_REMOVE(p, p_list);
432 	sx_xunlock(&allproc_lock);
433 
434 	sx_xlock(&proctree_lock);
435 	PROC_LOCK(p);
436 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
437 	PROC_UNLOCK(p);
438 
439 	/*
440 	 * killjobc() might drop and re-acquire proctree_lock to
441 	 * revoke control tty if exiting process was a session leader.
442 	 */
443 	killjobc();
444 
445 	/*
446 	 * Reparent all children processes:
447 	 * - traced ones to the original parent (or init if we are that parent)
448 	 * - the rest to init
449 	 */
450 	q = LIST_FIRST(&p->p_children);
451 	if (q != NULL)		/* only need this if any child is S_ZOMB */
452 		wakeup(q->p_reaper);
453 	for (; q != NULL; q = nq) {
454 		nq = LIST_NEXT(q, p_sibling);
455 		ksi = ksiginfo_alloc(TRUE);
456 		PROC_LOCK(q);
457 		q->p_sigparent = SIGCHLD;
458 
459 		if ((q->p_flag & P_TRACED) == 0) {
460 			proc_reparent(q, q->p_reaper, true);
461 			if (q->p_state == PRS_ZOMBIE) {
462 				/*
463 				 * Inform reaper about the reparented
464 				 * zombie, since wait(2) has something
465 				 * new to report.  Guarantee queueing
466 				 * of the SIGCHLD signal, similar to
467 				 * the _exit() behaviour, by providing
468 				 * our ksiginfo.  Ksi is freed by the
469 				 * signal delivery.
470 				 */
471 				if (q->p_ksi == NULL) {
472 					ksi1 = NULL;
473 				} else {
474 					ksiginfo_copy(q->p_ksi, ksi);
475 					ksi->ksi_flags |= KSI_INS;
476 					ksi1 = ksi;
477 					ksi = NULL;
478 				}
479 				PROC_LOCK(q->p_reaper);
480 				pksignal(q->p_reaper, SIGCHLD, ksi1);
481 				PROC_UNLOCK(q->p_reaper);
482 			} else if (q->p_pdeathsig > 0) {
483 				/*
484 				 * The child asked to received a signal
485 				 * when we exit.
486 				 */
487 				kern_psignal(q, q->p_pdeathsig);
488 			}
489 		} else {
490 			/*
491 			 * Traced processes are killed since their existence
492 			 * means someone is screwing up.
493 			 */
494 			t = proc_realparent(q);
495 			if (t == p) {
496 				proc_reparent(q, q->p_reaper, true);
497 			} else {
498 				PROC_LOCK(t);
499 				proc_reparent(q, t, true);
500 				PROC_UNLOCK(t);
501 			}
502 			/*
503 			 * Since q was found on our children list, the
504 			 * proc_reparent() call moved q to the orphan
505 			 * list due to present P_TRACED flag. Clear
506 			 * orphan link for q now while q is locked.
507 			 */
508 			proc_clear_orphan(q);
509 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
510 			q->p_flag2 &= ~P2_PTRACE_FSTP;
511 			q->p_ptevents = 0;
512 			FOREACH_THREAD_IN_PROC(q, tdt) {
513 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
514 				    TDB_FSTP);
515 			}
516 			kern_psignal(q, SIGKILL);
517 		}
518 		PROC_UNLOCK(q);
519 		if (ksi != NULL)
520 			ksiginfo_free(ksi);
521 	}
522 
523 	/*
524 	 * Also get rid of our orphans.
525 	 */
526 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
527 		PROC_LOCK(q);
528 		KASSERT(q->p_oppid == p->p_pid,
529 		    ("orphan %p of %p has unexpected oppid %d", q, p,
530 		    q->p_oppid));
531 		q->p_oppid = q->p_reaper->p_pid;
532 
533 		/*
534 		 * If we are the real parent of this process
535 		 * but it has been reparented to a debugger, then
536 		 * check if it asked for a signal when we exit.
537 		 */
538 		if (q->p_pdeathsig > 0)
539 			kern_psignal(q, q->p_pdeathsig);
540 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
541 		    q->p_pid);
542 		proc_clear_orphan(q);
543 		PROC_UNLOCK(q);
544 	}
545 
546 #ifdef KDTRACE_HOOKS
547 	if (SDT_PROBES_ENABLED()) {
548 		int reason = CLD_EXITED;
549 		if (WCOREDUMP(signo))
550 			reason = CLD_DUMPED;
551 		else if (WIFSIGNALED(signo))
552 			reason = CLD_KILLED;
553 		SDT_PROBE1(proc, , , exit, reason);
554 	}
555 #endif
556 
557 	/* Save exit status. */
558 	PROC_LOCK(p);
559 	p->p_xthread = td;
560 
561 #ifdef KDTRACE_HOOKS
562 	/*
563 	 * Tell the DTrace fasttrap provider about the exit if it
564 	 * has declared an interest.
565 	 */
566 	if (dtrace_fasttrap_exit)
567 		dtrace_fasttrap_exit(p);
568 #endif
569 
570 	/*
571 	 * Notify interested parties of our demise.
572 	 */
573 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
574 
575 	/*
576 	 * If this is a process with a descriptor, we may not need to deliver
577 	 * a signal to the parent.  proctree_lock is held over
578 	 * procdesc_exit() to serialize concurrent calls to close() and
579 	 * exit().
580 	 */
581 	signal_parent = 0;
582 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
583 		/*
584 		 * Notify parent that we're gone.  If parent has the
585 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
586 		 * notify process 1 instead (and hope it will handle this
587 		 * situation).
588 		 */
589 		PROC_LOCK(p->p_pptr);
590 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
591 		if (p->p_pptr->p_sigacts->ps_flag &
592 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
593 			struct proc *pp;
594 
595 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
596 			pp = p->p_pptr;
597 			PROC_UNLOCK(pp);
598 			proc_reparent(p, p->p_reaper, true);
599 			p->p_sigparent = SIGCHLD;
600 			PROC_LOCK(p->p_pptr);
601 
602 			/*
603 			 * Notify parent, so in case he was wait(2)ing or
604 			 * executing waitpid(2) with our pid, he will
605 			 * continue.
606 			 */
607 			wakeup(pp);
608 		} else
609 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
610 
611 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
612 			signal_parent = 1;
613 		} else if (p->p_sigparent != 0) {
614 			if (p->p_sigparent == SIGCHLD) {
615 				signal_parent = 1;
616 			} else { /* LINUX thread */
617 				signal_parent = 2;
618 			}
619 		}
620 	} else
621 		PROC_LOCK(p->p_pptr);
622 	sx_xunlock(&proctree_lock);
623 
624 	if (signal_parent == 1) {
625 		childproc_exited(p);
626 	} else if (signal_parent == 2) {
627 		kern_psignal(p->p_pptr, p->p_sigparent);
628 	}
629 
630 	/* Tell the prison that we are gone. */
631 	prison_proc_free(p->p_ucred->cr_prison);
632 
633 	/*
634 	 * The state PRS_ZOMBIE prevents other proesses from sending
635 	 * signal to the process, to avoid memory leak, we free memory
636 	 * for signal queue at the time when the state is set.
637 	 */
638 	sigqueue_flush(&p->p_sigqueue);
639 	sigqueue_flush(&td->td_sigqueue);
640 
641 	/*
642 	 * We have to wait until after acquiring all locks before
643 	 * changing p_state.  We need to avoid all possible context
644 	 * switches (including ones from blocking on a mutex) while
645 	 * marked as a zombie.  We also have to set the zombie state
646 	 * before we release the parent process' proc lock to avoid
647 	 * a lost wakeup.  So, we first call wakeup, then we grab the
648 	 * sched lock, update the state, and release the parent process'
649 	 * proc lock.
650 	 */
651 	wakeup(p->p_pptr);
652 	cv_broadcast(&p->p_pwait);
653 	sched_exit(p->p_pptr, td);
654 	PROC_SLOCK(p);
655 	p->p_state = PRS_ZOMBIE;
656 	PROC_UNLOCK(p->p_pptr);
657 
658 	/*
659 	 * Save our children's rusage information in our exit rusage.
660 	 */
661 	PROC_STATLOCK(p);
662 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
663 	PROC_STATUNLOCK(p);
664 
665 	/*
666 	 * Make sure the scheduler takes this thread out of its tables etc.
667 	 * This will also release this thread's reference to the ucred.
668 	 * Other thread parts to release include pcb bits and such.
669 	 */
670 	thread_exit();
671 }
672 
673 #ifndef _SYS_SYSPROTO_H_
674 struct abort2_args {
675 	char *why;
676 	int nargs;
677 	void **args;
678 };
679 #endif
680 
681 int
682 sys_abort2(struct thread *td, struct abort2_args *uap)
683 {
684 	struct proc *p = td->td_proc;
685 	struct sbuf *sb;
686 	void *uargs[16];
687 	int error, i, sig;
688 
689 	/*
690 	 * Do it right now so we can log either proper call of abort2(), or
691 	 * note, that invalid argument was passed. 512 is big enough to
692 	 * handle 16 arguments' descriptions with additional comments.
693 	 */
694 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
695 	sbuf_clear(sb);
696 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
697 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
698 	/*
699 	 * Since we can't return from abort2(), send SIGKILL in cases, where
700 	 * abort2() was called improperly
701 	 */
702 	sig = SIGKILL;
703 	/* Prevent from DoSes from user-space. */
704 	if (uap->nargs < 0 || uap->nargs > 16)
705 		goto out;
706 	if (uap->nargs > 0) {
707 		if (uap->args == NULL)
708 			goto out;
709 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
710 		if (error != 0)
711 			goto out;
712 	}
713 	/*
714 	 * Limit size of 'reason' string to 128. Will fit even when
715 	 * maximal number of arguments was chosen to be logged.
716 	 */
717 	if (uap->why != NULL) {
718 		error = sbuf_copyin(sb, uap->why, 128);
719 		if (error < 0)
720 			goto out;
721 	} else {
722 		sbuf_printf(sb, "(null)");
723 	}
724 	if (uap->nargs > 0) {
725 		sbuf_printf(sb, "(");
726 		for (i = 0;i < uap->nargs; i++)
727 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
728 		sbuf_printf(sb, ")");
729 	}
730 	/*
731 	 * Final stage: arguments were proper, string has been
732 	 * successfully copied from userspace, and copying pointers
733 	 * from user-space succeed.
734 	 */
735 	sig = SIGABRT;
736 out:
737 	if (sig == SIGKILL) {
738 		sbuf_trim(sb);
739 		sbuf_printf(sb, " (Reason text inaccessible)");
740 	}
741 	sbuf_cat(sb, "\n");
742 	sbuf_finish(sb);
743 	log(LOG_INFO, "%s", sbuf_data(sb));
744 	sbuf_delete(sb);
745 	exit1(td, 0, sig);
746 	return (0);
747 }
748 
749 #ifdef COMPAT_43
750 /*
751  * The dirty work is handled by kern_wait().
752  */
753 int
754 owait(struct thread *td, struct owait_args *uap __unused)
755 {
756 	int error, status;
757 
758 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
759 	if (error == 0)
760 		td->td_retval[1] = status;
761 	return (error);
762 }
763 #endif /* COMPAT_43 */
764 
765 /*
766  * The dirty work is handled by kern_wait().
767  */
768 int
769 sys_wait4(struct thread *td, struct wait4_args *uap)
770 {
771 	struct rusage ru, *rup;
772 	int error, status;
773 
774 	if (uap->rusage != NULL)
775 		rup = &ru;
776 	else
777 		rup = NULL;
778 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
779 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
780 		error = copyout(&status, uap->status, sizeof(status));
781 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
782 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
783 	return (error);
784 }
785 
786 int
787 sys_wait6(struct thread *td, struct wait6_args *uap)
788 {
789 	struct __wrusage wru, *wrup;
790 	siginfo_t si, *sip;
791 	idtype_t idtype;
792 	id_t id;
793 	int error, status;
794 
795 	idtype = uap->idtype;
796 	id = uap->id;
797 
798 	if (uap->wrusage != NULL)
799 		wrup = &wru;
800 	else
801 		wrup = NULL;
802 
803 	if (uap->info != NULL) {
804 		sip = &si;
805 		bzero(sip, sizeof(*sip));
806 	} else
807 		sip = NULL;
808 
809 	/*
810 	 *  We expect all callers of wait6() to know about WEXITED and
811 	 *  WTRAPPED.
812 	 */
813 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
814 
815 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
816 		error = copyout(&status, uap->status, sizeof(status));
817 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
818 		error = copyout(&wru, uap->wrusage, sizeof(wru));
819 	if (uap->info != NULL && error == 0)
820 		error = copyout(&si, uap->info, sizeof(si));
821 	return (error);
822 }
823 
824 /*
825  * Reap the remains of a zombie process and optionally return status and
826  * rusage.  Asserts and will release both the proctree_lock and the process
827  * lock as part of its work.
828  */
829 void
830 proc_reap(struct thread *td, struct proc *p, int *status, int options)
831 {
832 	struct proc *q, *t;
833 
834 	sx_assert(&proctree_lock, SA_XLOCKED);
835 	PROC_LOCK_ASSERT(p, MA_OWNED);
836 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
837 
838 	mtx_spin_wait_unlocked(&p->p_slock);
839 
840 	q = td->td_proc;
841 
842 	if (status)
843 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
844 	if (options & WNOWAIT) {
845 		/*
846 		 *  Only poll, returning the status.  Caller does not wish to
847 		 * release the proc struct just yet.
848 		 */
849 		PROC_UNLOCK(p);
850 		sx_xunlock(&proctree_lock);
851 		return;
852 	}
853 
854 	PROC_LOCK(q);
855 	sigqueue_take(p->p_ksi);
856 	PROC_UNLOCK(q);
857 
858 	/*
859 	 * If we got the child via a ptrace 'attach', we need to give it back
860 	 * to the old parent.
861 	 */
862 	if (p->p_oppid != p->p_pptr->p_pid) {
863 		PROC_UNLOCK(p);
864 		t = proc_realparent(p);
865 		PROC_LOCK(t);
866 		PROC_LOCK(p);
867 		CTR2(KTR_PTRACE,
868 		    "wait: traced child %d moved back to parent %d", p->p_pid,
869 		    t->p_pid);
870 		proc_reparent(p, t, false);
871 		PROC_UNLOCK(p);
872 		pksignal(t, SIGCHLD, p->p_ksi);
873 		wakeup(t);
874 		cv_broadcast(&p->p_pwait);
875 		PROC_UNLOCK(t);
876 		sx_xunlock(&proctree_lock);
877 		return;
878 	}
879 	PROC_UNLOCK(p);
880 
881 	/*
882 	 * Remove other references to this process to ensure we have an
883 	 * exclusive reference.
884 	 */
885 	sx_xlock(PIDHASHLOCK(p->p_pid));
886 	LIST_REMOVE(p, p_hash);
887 	sx_xunlock(PIDHASHLOCK(p->p_pid));
888 	LIST_REMOVE(p, p_sibling);
889 	reaper_abandon_children(p, true);
890 	reaper_clear(p);
891 	PROC_LOCK(p);
892 	proc_clear_orphan(p);
893 	PROC_UNLOCK(p);
894 	leavepgrp(p);
895 	if (p->p_procdesc != NULL)
896 		procdesc_reap(p);
897 	sx_xunlock(&proctree_lock);
898 
899 	proc_id_clear(PROC_ID_PID, p->p_pid);
900 
901 	PROC_LOCK(p);
902 	knlist_detach(p->p_klist);
903 	p->p_klist = NULL;
904 	PROC_UNLOCK(p);
905 
906 	/*
907 	 * Removal from allproc list and process group list paired with
908 	 * PROC_LOCK which was executed during that time should guarantee
909 	 * nothing can reach this process anymore. As such further locking
910 	 * is unnecessary.
911 	 */
912 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
913 
914 	PROC_LOCK(q);
915 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
916 	PROC_UNLOCK(q);
917 
918 	/*
919 	 * Decrement the count of procs running with this uid.
920 	 */
921 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
922 
923 	/*
924 	 * Destroy resource accounting information associated with the process.
925 	 */
926 #ifdef RACCT
927 	if (racct_enable) {
928 		PROC_LOCK(p);
929 		racct_sub(p, RACCT_NPROC, 1);
930 		PROC_UNLOCK(p);
931 	}
932 #endif
933 	racct_proc_exit(p);
934 
935 	/*
936 	 * Free credentials, arguments, and sigacts.
937 	 */
938 	proc_unset_cred(p);
939 	pargs_drop(p->p_args);
940 	p->p_args = NULL;
941 	sigacts_free(p->p_sigacts);
942 	p->p_sigacts = NULL;
943 
944 	/*
945 	 * Do any thread-system specific cleanups.
946 	 */
947 	thread_wait(p);
948 
949 	/*
950 	 * Give vm and machine-dependent layer a chance to free anything that
951 	 * cpu_exit couldn't release while still running in process context.
952 	 */
953 	vm_waitproc(p);
954 #ifdef MAC
955 	mac_proc_destroy(p);
956 #endif
957 
958 	KASSERT(FIRST_THREAD_IN_PROC(p),
959 	    ("proc_reap: no residual thread!"));
960 	uma_zfree(proc_zone, p);
961 	atomic_add_int(&nprocs, -1);
962 }
963 
964 static int
965 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
966     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
967     int check_only)
968 {
969 	struct rusage *rup;
970 
971 	sx_assert(&proctree_lock, SA_XLOCKED);
972 
973 	PROC_LOCK(p);
974 
975 	switch (idtype) {
976 	case P_ALL:
977 		if (p->p_procdesc == NULL ||
978 		   (p->p_pptr == td->td_proc &&
979 		   (p->p_flag & P_TRACED) != 0)) {
980 			break;
981 		}
982 
983 		PROC_UNLOCK(p);
984 		return (0);
985 	case P_PID:
986 		if (p->p_pid != (pid_t)id) {
987 			PROC_UNLOCK(p);
988 			return (0);
989 		}
990 		break;
991 	case P_PGID:
992 		if (p->p_pgid != (pid_t)id) {
993 			PROC_UNLOCK(p);
994 			return (0);
995 		}
996 		break;
997 	case P_SID:
998 		if (p->p_session->s_sid != (pid_t)id) {
999 			PROC_UNLOCK(p);
1000 			return (0);
1001 		}
1002 		break;
1003 	case P_UID:
1004 		if (p->p_ucred->cr_uid != (uid_t)id) {
1005 			PROC_UNLOCK(p);
1006 			return (0);
1007 		}
1008 		break;
1009 	case P_GID:
1010 		if (p->p_ucred->cr_gid != (gid_t)id) {
1011 			PROC_UNLOCK(p);
1012 			return (0);
1013 		}
1014 		break;
1015 	case P_JAILID:
1016 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1017 			PROC_UNLOCK(p);
1018 			return (0);
1019 		}
1020 		break;
1021 	/*
1022 	 * It seems that the thread structures get zeroed out
1023 	 * at process exit.  This makes it impossible to
1024 	 * support P_SETID, P_CID or P_CPUID.
1025 	 */
1026 	default:
1027 		PROC_UNLOCK(p);
1028 		return (0);
1029 	}
1030 
1031 	if (p_canwait(td, p)) {
1032 		PROC_UNLOCK(p);
1033 		return (0);
1034 	}
1035 
1036 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1037 		PROC_UNLOCK(p);
1038 		return (0);
1039 	}
1040 
1041 	/*
1042 	 * This special case handles a kthread spawned by linux_clone
1043 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1044 	 * functions need to be able to distinguish between waiting
1045 	 * on a process and waiting on a thread.  It is a thread if
1046 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1047 	 * signifies we want to wait for threads and not processes.
1048 	 */
1049 	if ((p->p_sigparent != SIGCHLD) ^
1050 	    ((options & WLINUXCLONE) != 0)) {
1051 		PROC_UNLOCK(p);
1052 		return (0);
1053 	}
1054 
1055 	if (siginfo != NULL) {
1056 		bzero(siginfo, sizeof(*siginfo));
1057 		siginfo->si_errno = 0;
1058 
1059 		/*
1060 		 * SUSv4 requires that the si_signo value is always
1061 		 * SIGCHLD. Obey it despite the rfork(2) interface
1062 		 * allows to request other signal for child exit
1063 		 * notification.
1064 		 */
1065 		siginfo->si_signo = SIGCHLD;
1066 
1067 		/*
1068 		 *  This is still a rough estimate.  We will fix the
1069 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1070 		 */
1071 		if (WCOREDUMP(p->p_xsig)) {
1072 			siginfo->si_code = CLD_DUMPED;
1073 			siginfo->si_status = WTERMSIG(p->p_xsig);
1074 		} else if (WIFSIGNALED(p->p_xsig)) {
1075 			siginfo->si_code = CLD_KILLED;
1076 			siginfo->si_status = WTERMSIG(p->p_xsig);
1077 		} else {
1078 			siginfo->si_code = CLD_EXITED;
1079 			siginfo->si_status = p->p_xexit;
1080 		}
1081 
1082 		siginfo->si_pid = p->p_pid;
1083 		siginfo->si_uid = p->p_ucred->cr_uid;
1084 
1085 		/*
1086 		 * The si_addr field would be useful additional
1087 		 * detail, but apparently the PC value may be lost
1088 		 * when we reach this point.  bzero() above sets
1089 		 * siginfo->si_addr to NULL.
1090 		 */
1091 	}
1092 
1093 	/*
1094 	 * There should be no reason to limit resources usage info to
1095 	 * exited processes only.  A snapshot about any resources used
1096 	 * by a stopped process may be exactly what is needed.
1097 	 */
1098 	if (wrusage != NULL) {
1099 		rup = &wrusage->wru_self;
1100 		*rup = p->p_ru;
1101 		PROC_STATLOCK(p);
1102 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1103 		PROC_STATUNLOCK(p);
1104 
1105 		rup = &wrusage->wru_children;
1106 		*rup = p->p_stats->p_cru;
1107 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1108 	}
1109 
1110 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1111 		proc_reap(td, p, status, options);
1112 		return (-1);
1113 	}
1114 	return (1);
1115 }
1116 
1117 int
1118 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1119     struct rusage *rusage)
1120 {
1121 	struct __wrusage wru, *wrup;
1122 	idtype_t idtype;
1123 	id_t id;
1124 	int ret;
1125 
1126 	/*
1127 	 * Translate the special pid values into the (idtype, pid)
1128 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1129 	 * kern_wait6() on its own.
1130 	 */
1131 	if (pid == WAIT_ANY) {
1132 		idtype = P_ALL;
1133 		id = 0;
1134 	} else if (pid < 0) {
1135 		idtype = P_PGID;
1136 		id = (id_t)-pid;
1137 	} else {
1138 		idtype = P_PID;
1139 		id = (id_t)pid;
1140 	}
1141 
1142 	if (rusage != NULL)
1143 		wrup = &wru;
1144 	else
1145 		wrup = NULL;
1146 
1147 	/*
1148 	 * For backward compatibility we implicitly add flags WEXITED
1149 	 * and WTRAPPED here.
1150 	 */
1151 	options |= WEXITED | WTRAPPED;
1152 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1153 	if (rusage != NULL)
1154 		*rusage = wru.wru_self;
1155 	return (ret);
1156 }
1157 
1158 static void
1159 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1160     int *status, int options, int si_code)
1161 {
1162 	bool cont;
1163 
1164 	PROC_LOCK_ASSERT(p, MA_OWNED);
1165 	sx_assert(&proctree_lock, SA_XLOCKED);
1166 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1167 	    si_code == CLD_CONTINUED);
1168 
1169 	cont = si_code == CLD_CONTINUED;
1170 	if ((options & WNOWAIT) == 0) {
1171 		if (cont)
1172 			p->p_flag &= ~P_CONTINUED;
1173 		else
1174 			p->p_flag |= P_WAITED;
1175 		PROC_LOCK(td->td_proc);
1176 		sigqueue_take(p->p_ksi);
1177 		PROC_UNLOCK(td->td_proc);
1178 	}
1179 	sx_xunlock(&proctree_lock);
1180 	if (siginfo != NULL) {
1181 		siginfo->si_code = si_code;
1182 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1183 	}
1184 	if (status != NULL)
1185 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1186 	PROC_UNLOCK(p);
1187 	td->td_retval[0] = p->p_pid;
1188 }
1189 
1190 int
1191 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1192     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1193 {
1194 	struct proc *p, *q;
1195 	pid_t pid;
1196 	int error, nfound, ret;
1197 	bool report;
1198 
1199 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1200 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1201 	AUDIT_ARG_VALUE(options);
1202 
1203 	q = td->td_proc;
1204 
1205 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1206 		PROC_LOCK(q);
1207 		id = (id_t)q->p_pgid;
1208 		PROC_UNLOCK(q);
1209 		idtype = P_PGID;
1210 	}
1211 
1212 	/* If we don't know the option, just return. */
1213 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1214 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1215 		return (EINVAL);
1216 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1217 		/*
1218 		 * We will be unable to find any matching processes,
1219 		 * because there are no known events to look for.
1220 		 * Prefer to return error instead of blocking
1221 		 * indefinitely.
1222 		 */
1223 		return (EINVAL);
1224 	}
1225 
1226 loop:
1227 	if (q->p_flag & P_STATCHILD) {
1228 		PROC_LOCK(q);
1229 		q->p_flag &= ~P_STATCHILD;
1230 		PROC_UNLOCK(q);
1231 	}
1232 	sx_xlock(&proctree_lock);
1233 loop_locked:
1234 	nfound = 0;
1235 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1236 		pid = p->p_pid;
1237 		ret = proc_to_reap(td, p, idtype, id, status, options,
1238 		    wrusage, siginfo, 0);
1239 		if (ret == 0)
1240 			continue;
1241 		else if (ret != 1) {
1242 			td->td_retval[0] = pid;
1243 			return (0);
1244 		}
1245 
1246 		nfound++;
1247 		PROC_LOCK_ASSERT(p, MA_OWNED);
1248 
1249 		if ((options & WTRAPPED) != 0 &&
1250 		    (p->p_flag & P_TRACED) != 0) {
1251 			PROC_SLOCK(p);
1252 			report =
1253 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1254 			    p->p_suspcount == p->p_numthreads &&
1255 			    (p->p_flag & P_WAITED) == 0);
1256 			PROC_SUNLOCK(p);
1257 			if (report) {
1258 			CTR4(KTR_PTRACE,
1259 			    "wait: returning trapped pid %d status %#x "
1260 			    "(xstat %d) xthread %d",
1261 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1262 			    p->p_xthread != NULL ?
1263 			    p->p_xthread->td_tid : -1);
1264 				report_alive_proc(td, p, siginfo, status,
1265 				    options, CLD_TRAPPED);
1266 				return (0);
1267 			}
1268 		}
1269 		if ((options & WUNTRACED) != 0 &&
1270 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1271 			PROC_SLOCK(p);
1272 			report = (p->p_suspcount == p->p_numthreads &&
1273 			    ((p->p_flag & P_WAITED) == 0));
1274 			PROC_SUNLOCK(p);
1275 			if (report) {
1276 				report_alive_proc(td, p, siginfo, status,
1277 				    options, CLD_STOPPED);
1278 				return (0);
1279 			}
1280 		}
1281 		if ((options & WCONTINUED) != 0 &&
1282 		    (p->p_flag & P_CONTINUED) != 0) {
1283 			report_alive_proc(td, p, siginfo, status, options,
1284 			    CLD_CONTINUED);
1285 			return (0);
1286 		}
1287 		PROC_UNLOCK(p);
1288 	}
1289 
1290 	/*
1291 	 * Look in the orphans list too, to allow the parent to
1292 	 * collect it's child exit status even if child is being
1293 	 * debugged.
1294 	 *
1295 	 * Debugger detaches from the parent upon successful
1296 	 * switch-over from parent to child.  At this point due to
1297 	 * re-parenting the parent loses the child to debugger and a
1298 	 * wait4(2) call would report that it has no children to wait
1299 	 * for.  By maintaining a list of orphans we allow the parent
1300 	 * to successfully wait until the child becomes a zombie.
1301 	 */
1302 	if (nfound == 0) {
1303 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1304 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1305 			    NULL, NULL, 1);
1306 			if (ret != 0) {
1307 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1308 				    (int)td->td_retval[0]));
1309 				PROC_UNLOCK(p);
1310 				nfound++;
1311 				break;
1312 			}
1313 		}
1314 	}
1315 	if (nfound == 0) {
1316 		sx_xunlock(&proctree_lock);
1317 		return (ECHILD);
1318 	}
1319 	if (options & WNOHANG) {
1320 		sx_xunlock(&proctree_lock);
1321 		td->td_retval[0] = 0;
1322 		return (0);
1323 	}
1324 	PROC_LOCK(q);
1325 	if (q->p_flag & P_STATCHILD) {
1326 		q->p_flag &= ~P_STATCHILD;
1327 		PROC_UNLOCK(q);
1328 		goto loop_locked;
1329 	}
1330 	sx_xunlock(&proctree_lock);
1331 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1332 	if (error)
1333 		return (error);
1334 	goto loop;
1335 }
1336 
1337 void
1338 proc_add_orphan(struct proc *child, struct proc *parent)
1339 {
1340 
1341 	sx_assert(&proctree_lock, SX_XLOCKED);
1342 	KASSERT((child->p_flag & P_TRACED) != 0,
1343 	    ("proc_add_orphan: not traced"));
1344 
1345 	if (LIST_EMPTY(&parent->p_orphans)) {
1346 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1347 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1348 	} else {
1349 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1350 		    child, p_orphan);
1351 	}
1352 	child->p_treeflag |= P_TREE_ORPHANED;
1353 }
1354 
1355 /*
1356  * Make process 'parent' the new parent of process 'child'.
1357  * Must be called with an exclusive hold of proctree lock.
1358  */
1359 void
1360 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1361 {
1362 
1363 	sx_assert(&proctree_lock, SX_XLOCKED);
1364 	PROC_LOCK_ASSERT(child, MA_OWNED);
1365 	if (child->p_pptr == parent)
1366 		return;
1367 
1368 	PROC_LOCK(child->p_pptr);
1369 	sigqueue_take(child->p_ksi);
1370 	PROC_UNLOCK(child->p_pptr);
1371 	LIST_REMOVE(child, p_sibling);
1372 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1373 
1374 	proc_clear_orphan(child);
1375 	if ((child->p_flag & P_TRACED) != 0) {
1376 		proc_add_orphan(child, child->p_pptr);
1377 	}
1378 
1379 	child->p_pptr = parent;
1380 	if (set_oppid)
1381 		child->p_oppid = parent->p_pid;
1382 }
1383